What it is
A reusable audit engine for testing whether model conclusions remain informative under declared pressure.
Reliability / current public state
GIL
Governed Invariance Logic
Current stateMEASURED
- Problem
- Mechanism
- Evidence
- Value
- Roadmap
Why it exists
Good predictive scores can conceal unstable rankings, leakage and fragile descriptor stories.
How it works
Negative controls, perturbation ladders, repeats, holdouts and claim ceilings turn model behaviour into inspectable evidence.
What is demonstrated
Direct reproduction, locked Materials Project shear and bulk repeats, and a controlled scale expansion are recorded.
Current implementation
Runnable audit routes and evidence-bundle tooling.
Why it matters
Helps teams decide which conclusions deserve further simulation, testing or investment.
Follow the evidence
Reference material
- Direct reproduction, locked Materials Project shear and bulk repeats, and a controlled scale expansion are recorded.
- Measured
- Bounded internal and external-docket demonstrations; no external validation or causal proof.